[SOLVED] G.Skill Trident Z F4 4133 19 cas 2x8gb XMP profile problems

[Candan]

Hi Everyone!

i7-8700k on Aorus Z370 Gaming 7 @ 5.0GHz
2 x 8gb Trident Z F4-4133C19D-16GTZ - Manufactured Oct 2017.
XMP 4133MHz. Stable

Recent crashes determined memory issues on one stick. RMA'd one stick. Kept running XMP profile on one 8gb stick without issue.

RMA came back (manufactured Feb 2019) identical model#

Added new stick. Reset CMOS. Booted with optimized default settings to Windows without issue.

Rebooted. Loaded XMP profile. Won't post. Sometimes I get advised of a boot issue and able to enter BIOS. Sometimes the PC just post loops and I have to reset CMOS again.

Still working OK with old stick standalone. No joy on new stick (standalone and in different slots)

Now you know where I'm at today! 🆒

I don't wish to run this RAM at stock. Its too expensive for that! It's also on the mobo compatibility list.

As of now, I have dram at 1.44v (any lower and there's no post or bsod.) vccio 1.13v and vccsa 1.15v. 3,732MHz. Nice and stable.

Question time!

- I'm a bit concerned about that voltage for sustained periods, that it's high. How the heck do G.Skill get a 4133MHz profile to run at 1.35v ?

- I know I don't need more speed than 3700MHz than I am using, but do you think it would be even possible to get to the 4133MHz profile speeds again like before the RMA? Why won't the new stick do what the old one did?

- If I'm stuck at 3700MHz forever, then so beit, but then how do I get the voltage down and stay stable. Or is that the billion dollar question?

Thanks for any pointers! PS I've read all the sticky guides

 

[Darkbreeze]

RMA'd one stick.

THAT, is a problem. Memory comes in sets for a reason. When a manufacturer tries to replace only one stick out of a set, I call BS. We've seen this time and time again. Recently in fact on a high end 8 stick kit. They should have replaced ALL of the sticks, with a whole new, matched, tested set. Not just one stick. A new stick could have wildly different specs even with the same exact part number OR simply have been assembled using memory chips from a different batch of ICs, which can be all it takes for two sticks to not want to play nice together. Doesn't even have to be anything "wrong" with either of the sticks, just, not the same or don't play well together.

At JEDEC speeds it's usually not a problem. Going beyond that to OC configurations is where unmatched memory tends to amplify differences.

Here's an example of one memory part number, with three entirely different configurations. Supposed to all be the same, but are not, even though timings and voltage and part number are identical. None of these sticks would play nice with any of the others.

https://forums.tomshardware.com/threads/amd-ram-compatibility.3210050/post-19785792

 

[Candan]

You make a good point. To be honest I've not considered this. Knew for different models / manufacturers but not for the same item.

I've contacted G.Skill and asked them why they didn't advise me this when they gave me the RMA for one stick. And tried to get them to take the new one back and the old stick and replace both. We'll see. I've heard issues from these guys before...

Having said all that. The XMP profile (or anything close) I cannot get to even with ONLY the one stick installed. So my first questions from the OP are still open to answer! 👍

Any suggestions? Thanks!

 

[Darkbreeze]

Have you tried upping the VCCIO and system agent voltage? These, and maybe one other setting, have a direct relationship with running memory at very high speeds on Intel platforms ever since at least Skylake, maybe also Haswell, possibly even Ivy bridge, but for SURE they are fundamental settings for very high OC speeds on Kaby and Coffee. I needed to make adjustments to both to get my 3000mhz Trident Z sticks to run stable at 3200mhz WITH tighter timings than the advertised specs. For very high speed sticks like yours, it might be essential. Honestly I've not tried running anything that high speed on the last few gens but the principles are the same.

I'd really work on getting matched sticks as well. Doesn't solve the issue of not being able to even get one stick to work right though. But, with a slight increase in VCCIO and system agent, you MIGHT be able to drop the actual DRAM voltage a bit if you play around with it.

Information on this here.

https://www.tweaktown.com/guides/8481/coffee-lake-overclocking-guide/index5.html

 

[Candan]

"As of now, I have dram at 1.44v (any lower and there's no post or bsod.) vccio 1.13v and vccsa 1.15v. 3,732MHz. Nice and stable."

But I still can't get over 3600+ MHz (I know I don't need it, but I've paid for it!)

I'd like to know how the XMP profile is still 1.35v @ 4133 MHz. If I drop DRAMV to 1.43 I'm not stable or can't boot. 1.44v I'm OK. I can go no lower! Maybe increase vccio and vccsa even more. But already pushing these!!

 

[Darkbreeze]

I see. Yes, that is troublesome.

I'm assuming you have these installed in the A2 and B2 slots, yes? Because if not, you need to move them to those slots and then retest. Second and fourth slots away from the CPU socket.

I don't like sending people away from here, but there are some (A few anyhow, not as much as when Tradesman1 was active over there) few members over at the G.Skill forums if you haven't tried there yet that might have something useful to offer.

Personally I think it's due to the stick, and the mixed factor. In this case, mixed meaning not factory tested together and shipped as a tested together set, which as I said is almost always a problem with speeds and timings outside of JEDEC specifications.

I'd INSIST that they replace them with a matched set, not just one stick. If they were 2666mhz or lower, not that big a deal. Over that, at ANY "OC" speed, absolutely a factor.

Any sticks that didn't come from the exact same production run, and then were tested, may not even be made up of the same components or ICs as in the example I linked you to earlier showing three different memory modules, with wildly different compositions, all using the exact same part number which means that at some point (multiple points actually) they changed the composition and makeup of that model between production runs. Usually this is done due to what they have available that will still be capable of providing the same speed, timings and run at the same voltage, even though those sticks might not run with other sticks that have the same specs but different compositions.

 

[Candan]

Gigabyte/Aorus are calling them 4,3,2 and 1 slots of course 🙄 I like your description of 2nd and 4th slots away from CPU. Makes it clear. In my case, it's 1 and 3. I've tried 2 and 4 too with similar results.

Thanks again. I'll also check G.Skill's forums. Hear there thoughts. Either way, I'll report back my findings.

Everything you've said about the component swap outs based is totally true. Usually caused by reduced cost or availability issues with POs still needed filled. Having said that, it still can't explain why I cannot get close to 1.35v or 4133MHz with only the new stick installed?

 

[Darkbreeze]

You have the memory, seemingly, installed in the wrong slots. They should be installed, based on the population rules from YOUR motherboard manul, in slots 1 and 2, not in slots 1 and 3 or slots 2 and 4. The channel A and channel B indicators seen below are not the same channels as you think they are. Those actually apply to a different subset of rules. You want slots 1 and 2, which are STILL the second and fourth slots away from the CPU socket, as seen here.

What you see in the box to the right of the slots below ARE the population rules for your board. And they are the same as for every modern consumer platform since at least Ivy bridge, whether on Intel or AMD platforms, for dual channel motherboard configurations.

 

[Candan]

EEEEkkk, I'm sorry you dug all this out! I was seeing the book by memory, and it was 1 and 3 in my mind! I do in fact have my sticks installed in 2 and 1, and have also tried 4 and 3.

In other words, where you've so kindly highlighted!

 

[Darkbreeze]

That's ok. Better to be sure, no matter what. That way we aren't chasing any ghosts or overlooking what might have been a much easier problem to solve. It's never a waste of time to double check the basics, because the basics are often where the problems are found. In this case, that's obviously improbable.

So on to the next round.

I may have missed it, but I don't recall seeing any mention of what BIOS version you are on. If you are not on version F12, then I would immediately update to that version now and try again. BIOS updates are usually pretty critical in the area of memory speed and compatibility and in fact OFTEN faster memory kits do not work with early BIOS versions and support for those are only added as time goes by and faster kits are released. I guarantee when that board was new, on the first BIOS release, there were no consumer 4133mhz memory kits available to the general market yet. They might have been around and getting ready for release, but they weren't being sold yet and they for sure weren't supported by any Z370 motherboard at initial release.

Also, while it's unlikely since I think you are moderately past the target audience I wrote this for, you might take a look at my XMP troubleshooting guide to see if maybe anything there jumps out or is useful in helping to determine the problem OR refine your test procedure.

*Resolving memory problems and setting up XMP/DOCP/AMP profiles

 

[Candan]

Thanks. I'm going to spend some time looking at the XMP profiles to dig deeper.

I am using the latest F12e BIOS. I've tried older ones too. Don't forget, when I originally purchased this in November 17, the BIOS was about F5 at the time and fully supported both 2017 sticks at XMP 4133.

I also hope to hear back from G.Skill today on my friendly email I sent them on Monday evening!

 

[Candan]

G.Skill have confirmed that I can return both sticks. That's what I'll do and we'll see from there!

I won't close the thread until we get to the after review!!

 

[Darkbreeze]

Nice. That's the right thing for them to do. Hopefully that will cure the issue.

 

[Candan]

SOooooo.

2 new sticks arrived from Gskill. XMP Profile is working perfectly @ 4133MHz

So G.Skill, thanks for replacing the first stick with another defective one.

Happy I got things working again. Annoyed I had to pay postage twice.

Thanks for your suggestions Darkbreeze

 

[Darkbreeze]

Cool man. This is EXACTLY why I tell people to NEVER accept the replacement of just one stick. There did not necessarily have to be anything wrong with the replacement module in order for it to not work properly with the existing module. This is WHY they sell them in kits. So replacing them individually is stupid on their part, and they know it, but they get away with it often because most people won't question it.